You Ask, I Answer: Avoiding Generative AI, Part 2: Infrastructure

This is part two of how to avoid generative AI altogether. This is a five-part series. This is part two. Today we're going to talk about infrastructure and security. Because presumably, if you want to avoid generative AI entirely, yes, you could just live in a cabin in the woods with you know you're off the grid and out of civilization.

But presumably that's not realistic for a lot of people. So what are the things that you can do as a business or as a person to reduce your exposure to generative AI? When it comes to infrastructure, you're gonna need to provide it. Today, a lot of SaaS software as a service companies provide cloud-based options, aka you access in a browser or something like that. Basically, you're using somebody else's computer.

And when your date is on somebody else's computer, they can, whether they're permitted to or not, they can use your data with things like AI. So the precaution you have to take is how much of my infrastructure do I want to bring in-house to keep my data physically away from other people, right? If you just don't trust other people, and there's good reason not to trust other people, you have to bring stuff in-house. So, what are those things? Number one, your website.

Who hosts it? What are the terms of service on it? And how is it being protected? So you want to host your website on your own hardware, hardware that you control. Uh, and then you can you you obviously host that at a co-location center, but that physical box is yours.

Uh, no one else is allowed on it. It's not managed. The colox has the co-location center has no access to it. You want your website on your own hardware. You want your email services on your own servers.

You wanted a box somewhere that is running in some form of Linux that is AI-free, running something like PostFix or another email service, so that you are providing your own email. Because you can't use, if you want to avoid generative AI, you can't use Gmail, you can't use Hotmail, you can't use any of the web-based services. You've got to host and run your own email to keep that data out of other people's hands. If you want to use social media, you can't use most social media. So the one of the few social networks that does not Mastodon, but you have to run the Mastodon node.

You have to set it up, run it yourself, and protect it so that data does not leave your Mastodon node. And then you can invite your friends and colleagues to be on that node and say, hey, this is a no AI node. Our data is not going to leave this node because we can't trust where it's going to go. You may not use uh data on this node for training AI. So you make that clear to your users, you cannot use AI with this stuff.

You will probably want to have a private cloud storage server somewhere as well, a physical server that has a big old hard disk that allows you to privately serve up your files. Because you can't use something like Dropbox or another service like that, because they all have the same terms of service that permit derivative works. And many of them, as far as I know, I don't think any of them carve out and say we will not use data you store with us uh for training AI models. If that derivative works clause is in there, and again, remember I'm not a lawyer, I can't give legal advice, but if that derivative works clause is in there, it means that legally they can use your data for training AI. So private cloud storage, your server hosted in a data center somewhere that you maintain that provides access to cloud services, cloud storage.

On the networking side, you want robust protection, you want firewalls that just block and knock away anybody that's not supposed to be there. Absolutely no uh ingress on your network. You want VPNs for you and your employees again to ensure that data transmission is secure in transit, that nobody's inspecting your data as it goes through their systems, so it's tunneled through their systems. There are some services that will have like double or triple tunneling or tunneling with encryption, whatever works best for you and whatever you can afford, you're gonna want to do that. You want tools, there's a bunch of tools.

There's one from the Mac called Lulu that allows you to inspect network traffic. So when you turn this thing on, you have to essentially configure it app by app to say, what is this app allowed to transmit back? So if you have an app that's supposedly local that is not talking to you know the cloud somewhere, and you turn on Lulu and Lulu says, Hey, it looks like this app's trying to communicate out. Do you want to allow this? You would say, Nope, not allowing it, and prohibit that app from communicating with the outside world.

On your website, people will say, you know, block crawlers with robots.txt. No, that does nothing. I mean, do it that just so that you have legal documentation. Like, yeah, I I I put this in robots.txt to explicitly declare, you know, you may not AI crawler is not welcome here. You want that in your content itself saying AI crawler is not welcome here.

You may you may not use my data to train AI. But you want to, if you really, really, really don't want your data ending up in an AI model, it cannot be on the public web, at least not unprotected. If you go on the dark web, there are plenty of websites where you have to enter a capture just to get to the website. Like you can't even see the website until you pass the uh the passcode test at the front, you know, uh drag the you know, tell you how many squares contain bicycles or whatever. That is the only way to stop your data from ending up in AI models is to have none of it available publicly.

You ha someone has to authenticate as human first. Because a lot of websites have things like RSS feeds to make blog readers easier. Well, an RSS feed is something that a crawler can find and extract data from it. So you can't have those. You can't allow RSS feeds.

Robots.txt does not stop a scraper that refuses to acknowledge it. If you write a scraper in uh a language like Python using Selenium uh or requests or scrapie, it's an option as to whether it obeys robots.txt or not. So you could say, nope, uh uh and you write a scraper that just ignores that entirely. There's no technical block to using that. That's why you've got to put a wall around all your content and watermark the crap out of it.

Put a wall around your content so that scraperbots can't get at it. I know, for example, any time that uh I publish a new blog post, I see versions of it on other websites, you know, some random blog spot bloggers copying it because they're scraping the RSS feed. It's one of the reasons why I put so many ads in my RSS feed, so that people are scraping my ads and putting them all over the internet. But if you don't want that to happen, you can't offer that, and you've got to lock things down so that nothing is publicly visible. When any kind of scraper shows up, it it immediately hits the captcha and the capture tells it to huh, see ya.

Don't let the door hit you or the good lord split ya. Um that's part two on infrastructure and security. The things you need to do from an infrastructure perspective if you want to prevent your data from ending up in generative AI models for people using your data to train their AI models. If you enjoyed this video, please hit the like button. Subscribe to my channel if you haven't already, and if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.